The NLEstimate macro allows you to estimate one or more linear or nonlinear combinations of parameters from any model for which you can save the model parameters and their variance-covariance matrix. Most modeling procedures which offer ESTIMATE, CONTRAST, or LSMEANS statements only provide for estimating or testing linear combinations of model parameters. However, common estimation problems often involve nonlinear combinations, particularly in generalized models with nonidentity link functions such as logistic and Poisson models.
This sample combines macro programming with PROC FREQ and DATA Step logic to count the number of missing and non-missing values for every variable in a data set. The results are stored in a data set.
This sample illustrates one method of counting the number of missing and non-missing values for each variable in a data set. Two methods for structuring the resulting data set are shown.
The %VARTEST macro provides a one-tailed test of the null hypothesis that the variance equals a non-zero constant for normally distributed data. It also provides point- and confidence interval estimates.
NOTE: The CIBASIC option in PROC UNIVARIATE provides one- and two-sided confidence intervals for the standard deviation and variance. PROC TTEST provides a confidence interval for the standard deviation using either of two methods.
PURPOSE:
The %VARTEST macro tests the null hypothesis that the variance (or standard deviation) of a set of independent and identically normally distributed values is equal to a specified constant against an alternative that the variance (or standard deviation) exceeds the constant. The macro also provides point- and confidence interval estimates for the variance and standard deviation.
NOTE: Beginning in SAS 9.2, the QIC statistic is produced by PROC GENMOD. Beginning in SAS 9.4 TS1M2, QIC is available in PROC GEE.
PURPOSE:
The %QIC macro computes the QIC and QICu statistics proposed by Pan (2001) for GEE (generalized estimating equations) models. These statistics allow comparisons of GEE models (model selection) and selection of a correlation structure.
The SELECT macro performs model selection methods for categorical-response models that can be fit in PROC LOGISTIC. These include models using the logit, probit, cloglog, cumulative logit, or generalized logit links. The macro supports binary as well as ordinal and nominal multinomial models.
Standard model selection is done by choosing candidate effects for entry to or removal from the model according to their significance levels. After completion, the set of models selected at each step of this process is sorted on the selected criterion - AUC, R-square, max-rescaled R-square, AIC, or BIC. The requested number of best models on the selected criterion is displayed.